When performing a test to see whether an estimator A for two populations is the same, is it more natural to use the difference

A1 – A2 = 0

or the ratio

A1 / A2 = 1

(or other)?

Is there a universal preference or is it situational?

Printable View

- May 7th 2017, 09:40 AMregineRatio vs difference when testing for equality?
When performing a test to see whether an estimator A for two populations is the same, is it more natural to use the difference

A1 – A2 = 0

or the ratio

A1 / A2 = 1

(or other)?

Is there a universal preference or is it situational? - May 7th 2017, 10:04 AMromsekRe: Ratio vs difference when testing for equality?
comparing means I'd use the expected value of the difference.

comparing standard deviations (or higher moments) I'd use the expected value of the ratio. - May 28th 2017, 02:17 AMregineRe: Ratio vs difference when testing for equality?
So would the logic "use difference for expectancy to account for potential expectancy = 0; use ratio everywhere else because it's generally easier."

Be correct?